Product taxonomy and SKU mapping for retail media

Nobody puts taxonomy at the front of a retail media pitch. Yet every audience definition, product set and sales report depends on it. If the catalog says one thing, the point-of-sale system says another and the advertiser uses a third category definition, the platform can deliver media while losing the ability to explain what it delivered against.

Product taxonomy is the structure that organizes products into categories and subcategories. SKU mapping connects each sellable item, pack, variant and identifier to the right product and category records. Together they form the translation layer between merchandise, transactions and media.

This guide covers the architecture: the hierarchy, identifiers, crosswalks, version control and governance that keep audiences and measurement comparable. Category affinity shows how governed product sets become an audience input.

Why a category tree is not enough

A category tree gives products a place to live. Retail media needs more.

A protein bar might sit under Snacks > Bars in one retailer, Health > Sports Nutrition in another and Confectionery > Functional Bars in a third. A brand may define its competitive set as all convenient protein products, including drinks and powders. None of those views is automatically wrong. The problem appears when teams move between them without a documented crosswalk.

GS1's Global Product Classification provides a shared hierarchy for grouping products (GS1 Global Product Classification). Retailers still need internal structures for navigation, ranging, financial reporting and local category management. A useful architecture does not replace every internal hierarchy with one global tree. It links them.

The minimum design usually has four layers:

1. Identity layer: the trade item and its identifiers.
2. Product layer: the underlying product family, brand, variant and pack relationships.
3. Classification layer: the primary merchandise hierarchy plus controlled facets.
4. Media layer: campaign groupings, competitive sets, promoted SKUs and halo sets.

Mixing those layers creates brittle data. A temporary campaign group should not silently rewrite the merchandise taxonomy. A new promotional bundle should not be treated as a new brand family simply because it has a new SKU.

SKU, product and product family

Teams often use "SKU" as if it means product. It usually means a retailer-specific stock-keeping record. The same underlying item may have different internal SKUs across countries, stores or systems. Different pack sizes need distinct trade-item identities but may roll into one product family for some analyses.

GS1 describes the GTIN as the identifier used to identify trade items (GS1 GTIN standard). GTINs do not solve every internal data problem, but they provide a stable external key when they are present and correctly managed.

A workable mapping model should distinguish:

• each sellable trade item;
• retailer SKU and store-level item codes;
• GTIN or other source identifiers;
• product family;
• variant attributes such as flavor or size;
• multipacks, bundles and promotional packs;
• replacement, discontinued and successor relationships;
• the taxonomy nodes and media groupings attached to each level.

Consider yogurt. A 150g strawberry cup, a 500g strawberry tub and a four-pack are not interchangeable at checkout. They may be separate SKUs and GTINs. They may still belong to one product family for brand reach, while pack-level analysis remains necessary for promoted-item sales and availability.

Crosswalks translate business views

A crosswalk maps a node or product set in one system to the corresponding nodes or products in another. Crosswalks are necessary because the retailer's category, the advertiser's portfolio and an external shopping feed rarely use identical structures.

Google's merchant system shows this dual structure: merchants can use Google's predefined category while also providing their own `product_type` hierarchy (Google product category guidance). Retail media needs the same explicitness. Store both values; do not overwrite one with the other.

Each crosswalk should record:

• source and target systems;
• source and target versions;
• relationship type: exact, broader, narrower or partial;
• included and excluded products;
• effective dates;
• owner and approval state;
• known exceptions.

A one-to-one table is not enough when the relationship is one-to-many. Suppose a brand's "functional beverages" set spans energy drinks, enhanced water, sports drinks and kombucha. The retailer may store those in four branches. The media layer can create a governed group across them without pretending that the merchandise hierarchy has changed.

Where taxonomy breaks targeting

Targeting fails when a requested audience and the catalog hierarchy do not describe the same product universe.

A brand asks for functional-beverage buyers. The platform selects only the retailer's energy-drink node. The audience is clean in technical terms but incomplete commercially. Another platform uses title keywords and includes protein powders. The audience is larger but noisy.

The brief should therefore bind the audience to a versioned product set. Record the category nodes, explicit inclusions, explicit exclusions and effective date. If the set changes during the campaign, preserve both versions and explain the effect on reach and outcome comparisons.

Taxonomy can also create false precision. An audience called "premium snack shoppers" may depend on a premium flag that is inconsistently populated. The label sounds useful; the input is not. Media buyers should ask for coverage and quality metrics on every attribute used to define eligibility.

Where taxonomy breaks measurement

Measurement fails when the denominator or product boundary changes.

A report claims category uplift for Dairy. Does Dairy include alternatives? Does it include cheese and butter when the campaign promoted yogurt? Were discontinued items removed? Did a taxonomy update move products between branches after launch?

IAB/MRC retail media guidance requires clear definitions for promoted products, brand halo and category halo, and stresses consistent reporting scope (IAB/MRC Retail Media Measurement Guidelines). The data architecture must make those definitions executable.

For every campaign, preserve:

• promoted SKUs;
• advertised product families;
• same-brand halo products;
• category halo products;
• competitive set, if used;
• excluded products;
• taxonomy and mapping version;
• store and time coverage;
• treatment of returns, substitutions and out-of-stocks.

Do not calculate halo by taking whatever currently sits under a category node. The campaign contract should define the halo set before results are reviewed.

Retailer taxonomy and brand taxonomy

Retailers organize products to run stores and ecommerce. Brands organize portfolios to manage categories and competition. A platform should not force either view to disappear.

Use a translation layer:

1. The retailer retains its primary merchandise hierarchy.
2. The platform maintains stable product identity and family relationships.
3. Advertiser views are represented as governed product sets or crosswalks.
4. Campaign definitions reference immutable versions of those sets.
5. Reports show which view produced each number.

This model also prevents a common political problem: arguing over whose taxonomy is "correct" when the actual need is to compare two legitimate business views.

Govern taxonomy changes

Taxonomy work is never finished. New products arrive, suppliers change descriptions, seasonal ranges appear and category strategy evolves. The valuable capability is not a perfect spreadsheet. It is a controlled change process.

A governance workflow should include:

• named data owners for identity, hierarchy and campaign mappings;
• change requests with business rationale;
• automated impact analysis before a node is moved or split;
• review of affected audiences, campaigns and historical reports;
• versioned publication with effective dates;
• rollback and correction procedures;
• service levels for unmapped and disputed products.

GS1 EPCIS is built around sharing visibility-event data with the what, when, where, why and how of products and assets (GS1 EPCIS standard). Retail media does not need to copy EPCIS architecture wholesale, but the provenance principle is useful: product data should carry enough context to show what changed, where it came from and when it became effective.

A minimum data model

A practical implementation can begin with the following tables or entities:

Trade item

Identifier, retailer SKU, GTIN where available, description, pack, variant, status and source.

Product family

Brand, range, parent product and relationships among sizes, flavors and packs.

Taxonomy node

Node ID, parent, name, definition, hierarchy version and effective dates.

Assignment

Trade item or family, node, assignment type, method, confidence, source and reviewer.

Crosswalk

Source node or product set, target node or set, relationship type, exceptions and version.

Campaign product set

Promoted, halo, competitor and excluded items bound to a campaign and date range.

The implementation technology matters less than preserving stable IDs, versions and lineage.

Validate the mapping layer before release

Run the same pre-release checks on every proposed taxonomy or crosswalk version:

• calculate the unmapped rate for active SKUs and list every commercially material gap;
• detect duplicate or conflicting identifiers for the same trade item;
• find orphan nodes, broken parent references and products assigned to retired nodes;
• review every unexpected one-to-many or many-to-many crosswalk rather than flattening it silently;
• reject effective-date gaps or overlaps that leave two versions active at once;
• reconcile promoted, halo, competitor and excluded campaign sets against the proposed mapping;
• reproduce one historical report using its original taxonomy and crosswalk version, then compare its product scope and totals with the archived output.

Publication should stop when an active product has no permitted mapping, identifiers conflict, a relationship has unexplained cardinality, effective dates overlap, a campaign set no longer reconciles or a historical report cannot be reproduced. Low-risk unmapped items may enter a documented exception queue only when they are excluded from affected audiences and reports.

Worked example: a protein-bar campaign

A retailer carries 80 bar SKUs. Its Snacks > Bars branch includes cereal bars, confectionery bars and protein bars. The advertiser defines its market as high-protein portable nutrition, which also includes two ready-to-drink products.

A defensible setup would:

1. resolve all 80 SKU identities and variants;
2. preserve the retailer's merchandise nodes;
3. create a governed protein-bar set based on classification rules;
4. create a wider advertiser market set that includes the two drinks;
5. freeze promoted, brand-halo and category-halo sets before launch;
6. report outcomes separately for promoted bars, same-brand halo and the wider market;
7. retain the set versions so a later category change does not rewrite history.

The point is not to build the largest audience. It is to make every inclusion reproducible.

Buyer checklist

Before approving a retail media data setup, ask:

• What identifies a trade item across systems?
• How are retailer SKUs connected to GTINs, product families and variants?
• Which hierarchy is authoritative for merchandise, media and reporting?
• How are external and advertiser taxonomies crosswalked?
• Can one-to-many and partial relationships be represented?
• Are promoted and halo sets frozen at campaign launch?
• Can historical reports be reproduced under the original version?
• What breaks when a taxonomy node changes?
• Who approves, publishes and reverses changes?

FAQ

Is a global taxonomy enough for retail media?

No. It provides a useful shared reference, but retailers still need local merchandise structures and advertisers need portfolio views. The systems need governed mappings.

Should pack sizes share one SKU?

No. Sellable packs generally need distinct item identities. They may roll into a shared product family for selected analyses.

What is the difference between a crosswalk and classification?

Classification assigns a product to a node. A crosswalk translates nodes or product sets between systems.

Why freeze campaign product sets?

Because taxonomy and assortment change. Freezing or versioning the set preserves the meaning of the campaign and makes later measurement reproducible.

Bottom line

Taxonomy and SKU mapping are not catalog housekeeping. They are the contract connecting product truth to audience delivery and sales proof. Keep identity, classification, crosswalks and campaign sets separate. Version every important relationship. Then targeting and reporting can survive the changes that happen in real retail systems.

Ready to see how this works in practice?

Footprints AI helps brands and retailers measure what matters. See our customer success stories or get in touch to discuss your retail media strategy.